人脸生成（Face Generation）¶

在该项目中，你将使用生成式对抗网络（Generative Adversarial Nets）来生成新的人脸图像。

获取数据¶

该项目将使用以下数据集：

MNIST
CelebA

由于 CelebA 数据集比较复杂，而且这是你第一次使用 GANs。我们想让你先在 MNIST 数据集上测试你的 GANs 模型，以让你更快的评估所建立模型的性能。

如果你在使用 FloydHub, 请将 data_dir 设置为 "/input" 并使用 FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".

data_dir = './data'

# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
#data_dir = '/input'


"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper

helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)

Found mnist Data
Found celeba Data

探索数据（Explore the Data）¶

MNIST¶

MNIST 是一个手写数字的图像数据集。你可以更改 show_n_images 探索此数据集。

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot

mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')

<matplotlib.image.AxesImage at 0x7f7ed2a6a400>

CelebA¶

CelebFaces Attributes Dataset (CelebA) 是一个包含 20 多万张名人图片及相关图片说明的数据集。你将用此数据集生成人脸，不会用不到相关说明。你可以更改 show_n_images 探索此数据集。

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))

<matplotlib.image.AxesImage at 0x7f7ed2991f60>

预处理数据（Preprocess the Data）¶

由于该项目的重点是建立 GANs 模型，我们将为你预处理数据。

经过数据预处理，MNIST 和 CelebA 数据集的值在 28×28 维度图像的 [-0.5, 0.5] 范围内。CelebA 数据集中的图像裁剪了非脸部的图像部分，然后调整到 28x28 维度。

MNIST 数据集中的图像是单通道的黑白图像，CelebA 数据集中的图像是三通道的 RGB 彩色图像。

建立神经网络（Build the Neural Network）¶

你将通过部署以下函数来建立 GANs 的主要组成部分:

model_inputs
discriminator
generator
model_loss
model_opt
train

检查 TensorFlow 版本并获取 GPU 型号¶

检查你是否使用正确的 TensorFlow 版本，并获取 GPU 型号

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf

# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer.  You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))

# Check for a GPU
if not tf.test.gpu_device_name():
    warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))

TensorFlow Version: 1.0.0
Default GPU Device: /gpu:0

输入（Input）¶

部署 model_inputs 函数以创建用于神经网络的占位符 (TF Placeholders)。请创建以下占位符：

输入图像占位符: 使用 image_width，image_height 和 image_channels 设置为 rank 4。
输入 Z 占位符: 设置为 rank 2，并命名为 z_dim。
学习速率占位符: 设置为 rank 0。

返回占位符元组的形状为 (tensor of real input images, tensor of z data, learning rate)。

import problem_unittests as tests

def model_inputs(image_width, image_height, image_channels, z_dim):
    """
    Create the model inputs
    :param image_width: The input image width
    :param image_height: The input image height
    :param image_channels: The number of image channels
    :param z_dim: The dimension of Z
    :return: Tuple of (tensor of real input images, tensor of z data, learning rate)
    """
    # TODO: Implement Function
    inputs_real = tf.placeholder(tf.float32, (None, image_width, image_height, image_channels), name='input_real')
    inputs_z = tf.placeholder(tf.float32, (None, z_dim), name='z_dim')
    learning_rate = tf.placeholder(tf.float32, name='learning_rate')

    return inputs_real, inputs_z, learning_rate


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)

Tests Passed

辨别器（Discriminator）¶

部署 discriminator 函数创建辨别器神经网络以辨别 images。该函数应能够重复使用神经网络中的各种变量。在 tf.variable_scope 中使用 "discriminator" 的变量空间名来重复使用该函数中的变量。

该函数应返回形如 (tensor output of the discriminator, tensor logits of the discriminator) 的元组。

def discriminator(images, reuse=False):
    """
    Create the discriminator network
    :param image: Tensor of input image(s)
    :param reuse: Boolean if the weights should be reused
    :return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
    """
    # TODO: Implement Function
    alpha = 0.2
    with tf.variable_scope('discriminator', reuse=reuse):
        # Input layer is 28*28*?
        #print(images.shape)
        x1 = tf.layers.conv2d(images, 64, 5, strides=2, padding='same')
        #print(x1.shape)
        x1 = tf.maximum(alpha * x1, x1)
        # 14*14*64
        
        x2 = tf.layers.conv2d(x1, 256, 5, strides=2, padding='same')
        #print(x2.shape)
        x2 = tf.layers.batch_normalization(x2, training=True)
        x2 = tf.maximum(alpha * x2, x2)
        # 7*7*256
        
#         x3 = tf.layers.conv2d(x2, 256, 5, strides=2, padding='same')
#         #print(x3.shape)
#         x3 = tf.layers.batch_normalization(x3, training=True)
#         x3 = tf.maximum(alpha * x3, x3)
#         # 4*4*256

        # Flatten it
        flat = tf.reshape(x2, (-1, 7*7*256))
        #print(flat.shape)
        logits = tf.layers.dense(flat, 1)
        out = tf.sigmoid(logits)
        
        return out, logits


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)

Tests Passed

生成器（Generator）¶

部署 generator 函数以使用 z 生成图像。该函数应能够重复使用神经网络中的各种变量。在 tf.variable_scope 中使用 "generator" 的变量空间名来重复使用该函数中的变量。

该函数应返回所生成的 28 x 28 x out_channel_dim 维度图像。

def generator(z, out_channel_dim, is_train=True):
    """
    Create the generator network
    :param z: Input z
    :param out_channel_dim: The number of channels in the output image
    :param is_train: Boolean if generator is being used for training
    :return: The tensor output of the generator
    """
    # TODO: Implement Function
    alpha = 0.2
    with tf.variable_scope('generator', reuse=not is_train):
        # First fully connected layer
        x1 = tf.layers.dense(z, 7*7*256)
        # Reshape it to start the convolutional stack
        x1 = tf.reshape(x1, (-1, 7, 7, 256))
        #print(x1.shape)
        x1 = tf.layers.batch_normalization(x1, training=is_train)
        x1 = tf.maximum(alpha * x1, x1)
        # now 7*7*256
        
#         x2 = tf.layers.conv2d_transpose(x1, 128, 5, strides=1, padding='valid')
#         #print(x2.shape)
#         x2 = tf.layers.batch_normalization(x2, training=is_train)
#         x2 = tf.maximum(alpha * x2, x2)
#         # 7*7*128 now
        
        x3 = tf.layers.conv2d_transpose(x1, 64, 5, strides=2, padding='same')
        #print(x3.shape)
        x3 = tf.layers.batch_normalization(x3, training=is_train)
        x3 = tf.maximum(alpha * x3, x3)
        # 14*14*64 now
        
        # Output layer
        logits = tf.layers.conv2d_transpose(x3, out_channel_dim, 5, strides=2, padding='same')
        #print(logits.shape)
        # 28*28*? now
        
        out = tf.tanh(logits)
        
        return out


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)

Tests Passed

损失函数（Loss）¶

部署 model_loss 函数训练并计算 GANs 的损失。该函数应返回形如 (discriminator loss, generator loss) 的元组。

使用你已实现的函数：

discriminator(images, reuse=False)
generator(z, out_channel_dim, is_train=True)

def model_loss(input_real, input_z, out_channel_dim):
    """
    Get the loss for the discriminator and generator
    :param input_real: Images from the real dataset
    :param input_z: Z input
    :param out_channel_dim: The number of channels in the output image
    :return: A tuple of (discriminator loss, generator loss)
    """
    # TODO: Implement Function
    g_model = generator(input_z, out_channel_dim)
    d_model_real, d_logits_real = discriminator(input_real)
    d_model_fake, d_logits_fake = discriminator(g_model, reuse=True)

    d_loss_real = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_real, labels=tf.ones_like(d_model_real)))
    d_loss_fake = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, labels=tf.zeros_like(d_model_fake)))
    g_loss = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, labels=tf.ones_like(d_model_fake)))

    d_loss = d_loss_real + d_loss_fake

    return d_loss, g_loss


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)

Tests Passed

优化（Optimization）¶

部署 model_opt 函数实现对 GANs 的优化。使用 tf.trainable_variables 获取可训练的所有变量。通过变量空间名 discriminator 和 generator 来过滤变量。该函数应返回形如 (discriminator training operation, generator training operation) 的元组。

def model_opt(d_loss, g_loss, learning_rate, beta1):
    """
    Get optimization operations
    :param d_loss: Discriminator loss Tensor
    :param g_loss: Generator loss Tensor
    :param learning_rate: Learning Rate Placeholder
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :return: A tuple of (discriminator training operation, generator training operation)
    """
    # TODO: Implement Function
    # Get weights and bias to update
    t_vars = tf.trainable_variables()
    d_vars = [var for var in t_vars if var.name.startswith('discriminator')]
    g_vars = [var for var in t_vars if var.name.startswith('generator')]

    # Optimize
    with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
        d_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(d_loss, var_list=d_vars)
        g_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(g_loss, var_list=g_vars)

    return d_train_opt, g_train_opt


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)

Tests Passed

训练神经网络（Neural Network Training）¶

输出显示¶

使用该函数可以显示生成器 (Generator) 在训练过程中的当前输出，这会帮你评估 GANs 模型的训练程度。

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np

def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
    """
    Show example output for the generator
    :param sess: TensorFlow session
    :param n_images: Number of Images to display
    :param input_z: Input Z Tensor
    :param out_channel_dim: The number of channels in the output image
    :param image_mode: The mode to use for images ("RGB" or "L")
    """
    cmap = None if image_mode == 'RGB' else 'gray'
    z_dim = input_z.get_shape().as_list()[-1]
    example_z = np.random.uniform(-1, 1, size=[n_images, z_dim])

    samples = sess.run(
        generator(input_z, out_channel_dim, False),
        feed_dict={input_z: example_z})

    images_grid = helper.images_square_grid(samples, image_mode)
    pyplot.imshow(images_grid, cmap=cmap)
    pyplot.show()

训练¶

部署 train 函数以建立并训练 GANs 模型。记得使用以下你已完成的函数：

model_inputs(image_width, image_height, image_channels, z_dim)
model_loss(input_real, input_z, out_channel_dim)
model_opt(d_loss, g_loss, learning_rate, beta1)

使用 show_generator_output 函数显示 generator 在训练过程中的输出。

注意：在每个批次 (batch) 中运行 show_generator_output 函数会显著增加训练时间与该 notebook 的体积。推荐每 100 批次输出一次 generator 的输出。

def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
    """
    Train the GAN
    :param epoch_count: Number of epochs
    :param batch_size: Batch Size
    :param z_dim: Z dimension
    :param learning_rate: Learning Rate
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :param get_batches: Function to get batches
    :param data_shape: Shape of the data
    :param data_image_mode: The image mode to use for images ("RGB" or "L")
    """
    # TODO: Build Model
    image_count, image_width, image_height, image_channels = data_shape
    input_real, input_z, tf_lr = model_inputs(image_width, image_height, image_channels, z_dim)
    d_loss, g_loss = model_loss(input_real, input_z, image_channels)
    d_opt, g_opt = model_opt(d_loss, g_loss, tf_lr, beta1)
    
    print_every = 10
    show_every = 100
    steps = 0
    total_batches = int(image_count/batch_size)+1
    
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for epoch_i in range(epoch_count):
            batch_i = 0
            for batch_images in get_batches(batch_size):
                #print(batch_images.shape) # => (128, 28, 28, 1)
                steps += 1
                batch_i += 1
                # TODO: Train Model
                # Sample random noise for G
                batch_z = np.random.uniform(-1, 1, size=(batch_size, z_dim))

                # Run optimizers
                _ = sess.run(d_opt, feed_dict={input_z: batch_z, input_real: batch_images, tf_lr: learning_rate})
                _ = sess.run(g_opt, feed_dict={input_z: batch_z, input_real: batch_images, tf_lr: learning_rate})
                
                if steps % print_every == 0:
                    # At the end of each epoch, get the losses and print them out
                    train_loss_d = d_loss.eval({input_z: batch_z, input_real: batch_images})
                    train_loss_g = g_loss.eval({input_z: batch_z})

                    print("Epoch {:>2}/{} Batch {:>4}/{}...".format(epoch_i+1, epoch_count, batch_i, total_batches),
                          "Discriminator Loss: {:.4f}...".format(train_loss_d),
                          "Generator Loss: {:.4f}".format(train_loss_g))
                if steps % show_every == 0:
                    show_generator_output(sess, show_n_images, input_z, image_channels, data_image_mode)
        # final show
        show_generator_output(sess, show_n_images, input_z, image_channels, data_image_mode)

MNIST¶

在 MNIST 上测试你的 GANs 模型。经过 2 次迭代，GANs 应该能够生成类似手写数字的图像。确保生成器 (generator) 低于辨别器 (discriminator) 的损失，或接近 0。

# mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
# print(mnist_dataset.shape) # => (60000, 28, 28, 1)
# print(mnist_dataset.image_mode) # => L

batch_size = 128
z_dim = 90
learning_rate = 0.002
beta1 = 0.5


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 2

mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
          mnist_dataset.shape, mnist_dataset.image_mode)

Epoch  1/2 Batch   10/469... Discriminator Loss: 8.5929... Generator Loss: 0.5849
Epoch  1/2 Batch   20/469... Discriminator Loss: 0.3658... Generator Loss: 3.3214
Epoch  1/2 Batch   30/469... Discriminator Loss: 6.1468... Generator Loss: 0.0492
Epoch  1/2 Batch   40/469... Discriminator Loss: 0.6277... Generator Loss: 2.0318
Epoch  1/2 Batch   50/469... Discriminator Loss: 2.6176... Generator Loss: 6.9106
Epoch  1/2 Batch   60/469... Discriminator Loss: 0.6704... Generator Loss: 3.0671
Epoch  1/2 Batch   70/469... Discriminator Loss: 0.2353... Generator Loss: 3.7157
Epoch  1/2 Batch   80/469... Discriminator Loss: 0.1221... Generator Loss: 3.1437
Epoch  1/2 Batch   90/469... Discriminator Loss: 0.1228... Generator Loss: 2.7797
Epoch  1/2 Batch  100/469... Discriminator Loss: 0.8252... Generator Loss: 0.8013

Epoch  1/2 Batch  110/469... Discriminator Loss: 3.4432... Generator Loss: 0.0990
Epoch  1/2 Batch  120/469... Discriminator Loss: 1.4417... Generator Loss: 0.4343
Epoch  1/2 Batch  130/469... Discriminator Loss: 0.6258... Generator Loss: 5.4060
Epoch  1/2 Batch  140/469... Discriminator Loss: 1.1804... Generator Loss: 5.3933
Epoch  1/2 Batch  150/469... Discriminator Loss: 1.8064... Generator Loss: 7.7730
Epoch  1/2 Batch  160/469... Discriminator Loss: 0.1071... Generator Loss: 5.0246
Epoch  1/2 Batch  170/469... Discriminator Loss: 0.2389... Generator Loss: 4.4654
Epoch  1/2 Batch  180/469... Discriminator Loss: 0.1797... Generator Loss: 4.2574
Epoch  1/2 Batch  190/469... Discriminator Loss: 0.1545... Generator Loss: 3.7126
Epoch  1/2 Batch  200/469... Discriminator Loss: 1.1187... Generator Loss: 1.2844

Epoch  1/2 Batch  210/469... Discriminator Loss: 1.1020... Generator Loss: 3.9345
Epoch  1/2 Batch  220/469... Discriminator Loss: 0.8305... Generator Loss: 1.0736
Epoch  1/2 Batch  230/469... Discriminator Loss: 1.0111... Generator Loss: 2.9242
Epoch  1/2 Batch  240/469... Discriminator Loss: 0.8744... Generator Loss: 1.1520
Epoch  1/2 Batch  250/469... Discriminator Loss: 1.1409... Generator Loss: 4.7553
Epoch  1/2 Batch  260/469... Discriminator Loss: 0.5528... Generator Loss: 2.6395
Epoch  1/2 Batch  270/469... Discriminator Loss: 1.5083... Generator Loss: 0.4426
Epoch  1/2 Batch  280/469... Discriminator Loss: 0.5942... Generator Loss: 2.0435
Epoch  1/2 Batch  290/469... Discriminator Loss: 0.8457... Generator Loss: 1.9918
Epoch  1/2 Batch  300/469... Discriminator Loss: 0.8518... Generator Loss: 0.8441

Epoch  1/2 Batch  310/469... Discriminator Loss: 0.5537... Generator Loss: 2.2694
Epoch  1/2 Batch  320/469... Discriminator Loss: 0.4213... Generator Loss: 2.4824
Epoch  1/2 Batch  330/469... Discriminator Loss: 1.2043... Generator Loss: 0.4693
Epoch  1/2 Batch  340/469... Discriminator Loss: 0.5925... Generator Loss: 1.0819
Epoch  1/2 Batch  350/469... Discriminator Loss: 3.7171... Generator Loss: 0.0628
Epoch  1/2 Batch  360/469... Discriminator Loss: 0.8511... Generator Loss: 0.9785
Epoch  1/2 Batch  370/469... Discriminator Loss: 0.4531... Generator Loss: 2.5551
Epoch  1/2 Batch  380/469... Discriminator Loss: 0.3613... Generator Loss: 1.8179
Epoch  1/2 Batch  390/469... Discriminator Loss: 0.5280... Generator Loss: 1.4455
Epoch  1/2 Batch  400/469... Discriminator Loss: 0.7917... Generator Loss: 0.7605

Epoch  1/2 Batch  410/469... Discriminator Loss: 2.3727... Generator Loss: 0.1784
Epoch  1/2 Batch  420/469... Discriminator Loss: 0.1228... Generator Loss: 3.0682
Epoch  1/2 Batch  430/469... Discriminator Loss: 2.1249... Generator Loss: 0.1876
Epoch  1/2 Batch  440/469... Discriminator Loss: 0.4203... Generator Loss: 1.3994
Epoch  1/2 Batch  450/469... Discriminator Loss: 1.0580... Generator Loss: 0.5849
Epoch  1/2 Batch  460/469... Discriminator Loss: 0.2167... Generator Loss: 2.4061
Epoch  2/2 Batch    2/469... Discriminator Loss: 0.1310... Generator Loss: 2.9959
Epoch  2/2 Batch   12/469... Discriminator Loss: 0.1955... Generator Loss: 2.1429
Epoch  2/2 Batch   22/469... Discriminator Loss: 1.0287... Generator Loss: 5.4971
Epoch  2/2 Batch   32/469... Discriminator Loss: 0.5562... Generator Loss: 1.1538

Epoch  2/2 Batch   42/469... Discriminator Loss: 0.5747... Generator Loss: 1.1693
Epoch  2/2 Batch   52/469... Discriminator Loss: 0.2870... Generator Loss: 2.4210
Epoch  2/2 Batch   62/469... Discriminator Loss: 0.2252... Generator Loss: 2.3610
Epoch  2/2 Batch   72/469... Discriminator Loss: 0.1814... Generator Loss: 3.0147
Epoch  2/2 Batch   82/469... Discriminator Loss: 0.1368... Generator Loss: 2.7805
Epoch  2/2 Batch   92/469... Discriminator Loss: 0.1586... Generator Loss: 2.6264
Epoch  2/2 Batch  102/469... Discriminator Loss: 0.1238... Generator Loss: 3.2263
Epoch  2/2 Batch  112/469... Discriminator Loss: 0.0723... Generator Loss: 3.8006
Epoch  2/2 Batch  122/469... Discriminator Loss: 0.4937... Generator Loss: 4.9834
Epoch  2/2 Batch  132/469... Discriminator Loss: 0.5329... Generator Loss: 1.4532

Epoch  2/2 Batch  142/469... Discriminator Loss: 1.2976... Generator Loss: 0.7672
Epoch  2/2 Batch  152/469... Discriminator Loss: 0.4248... Generator Loss: 2.1029
Epoch  2/2 Batch  162/469... Discriminator Loss: 0.3554... Generator Loss: 1.6209
Epoch  2/2 Batch  172/469... Discriminator Loss: 0.2499... Generator Loss: 2.0900
Epoch  2/2 Batch  182/469... Discriminator Loss: 0.2186... Generator Loss: 2.3867
Epoch  2/2 Batch  192/469... Discriminator Loss: 0.2678... Generator Loss: 2.0116
Epoch  2/2 Batch  202/469... Discriminator Loss: 0.2033... Generator Loss: 2.7920
Epoch  2/2 Batch  212/469... Discriminator Loss: 0.1637... Generator Loss: 2.6328
Epoch  2/2 Batch  222/469... Discriminator Loss: 0.1439... Generator Loss: 2.6231
Epoch  2/2 Batch  232/469... Discriminator Loss: 0.1483... Generator Loss: 2.5971

Epoch  2/2 Batch  242/469... Discriminator Loss: 0.1918... Generator Loss: 2.1754
Epoch  2/2 Batch  252/469... Discriminator Loss: 0.1392... Generator Loss: 3.0990
Epoch  2/2 Batch  262/469... Discriminator Loss: 0.1445... Generator Loss: 2.5714
Epoch  2/2 Batch  272/469... Discriminator Loss: 1.6170... Generator Loss: 2.3008
Epoch  2/2 Batch  282/469... Discriminator Loss: 1.9689... Generator Loss: 5.1268
Epoch  2/2 Batch  292/469... Discriminator Loss: 0.6605... Generator Loss: 1.2792
Epoch  2/2 Batch  302/469... Discriminator Loss: 1.0842... Generator Loss: 0.6482
Epoch  2/2 Batch  312/469... Discriminator Loss: 0.7603... Generator Loss: 0.8563
Epoch  2/2 Batch  322/469... Discriminator Loss: 0.4169... Generator Loss: 1.6791
Epoch  2/2 Batch  332/469... Discriminator Loss: 0.2775... Generator Loss: 2.0030

Epoch  2/2 Batch  342/469... Discriminator Loss: 0.2603... Generator Loss: 2.0885
Epoch  2/2 Batch  352/469... Discriminator Loss: 0.2569... Generator Loss: 2.0814
Epoch  2/2 Batch  362/469... Discriminator Loss: 0.2161... Generator Loss: 2.3699
Epoch  2/2 Batch  372/469... Discriminator Loss: 0.1312... Generator Loss: 2.8338
Epoch  2/2 Batch  382/469... Discriminator Loss: 0.1164... Generator Loss: 3.1886
Epoch  2/2 Batch  392/469... Discriminator Loss: 0.1716... Generator Loss: 2.3939
Epoch  2/2 Batch  402/469... Discriminator Loss: 0.1830... Generator Loss: 2.5341
Epoch  2/2 Batch  412/469... Discriminator Loss: 0.1588... Generator Loss: 2.6110
Epoch  2/2 Batch  422/469... Discriminator Loss: 0.1067... Generator Loss: 3.2803
Epoch  2/2 Batch  432/469... Discriminator Loss: 0.0821... Generator Loss: 3.6532

Epoch  2/2 Batch  442/469... Discriminator Loss: 0.1001... Generator Loss: 3.5203
Epoch  2/2 Batch  452/469... Discriminator Loss: 1.1622... Generator Loss: 3.5862
Epoch  2/2 Batch  462/469... Discriminator Loss: 2.8575... Generator Loss: 7.6113

CelebA¶

在 CelebA 上运行你的 GANs 模型。在一般的GPU上运行每次迭代大约需要 20 分钟。你可以运行整个迭代，或者当 GANs 开始产生真实人脸图像时停止它。

# celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
# print(celeba_dataset.shape) # => (202599, 28, 28, 3)
# print(celeba_dataset.image_mode) # => RGB

batch_size = 64
z_dim = 200
learning_rate = 0.001
beta1 = 0.5


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 5

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)

Epoch  1/5 Batch   10/3166... Discriminator Loss: 0.3102... Generator Loss: 3.2074
Epoch  1/5 Batch   20/3166... Discriminator Loss: 5.6237... Generator Loss: 0.2373
Epoch  1/5 Batch   30/3166... Discriminator Loss: 0.3167... Generator Loss: 13.3016
Epoch  1/5 Batch   40/3166... Discriminator Loss: 0.2502... Generator Loss: 3.8342
Epoch  1/5 Batch   50/3166... Discriminator Loss: 0.8028... Generator Loss: 0.9192
Epoch  1/5 Batch   60/3166... Discriminator Loss: 0.2187... Generator Loss: 2.8552
Epoch  1/5 Batch   70/3166... Discriminator Loss: 1.2090... Generator Loss: 0.5221
Epoch  1/5 Batch   80/3166... Discriminator Loss: 0.2594... Generator Loss: 9.1774
Epoch  1/5 Batch   90/3166... Discriminator Loss: 0.6313... Generator Loss: 0.9891
Epoch  1/5 Batch  100/3166... Discriminator Loss: 0.6758... Generator Loss: 8.8822

Epoch  1/5 Batch  110/3166... Discriminator Loss: 0.8531... Generator Loss: 0.7821
Epoch  1/5 Batch  120/3166... Discriminator Loss: 1.8129... Generator Loss: 0.2775
Epoch  1/5 Batch  130/3166... Discriminator Loss: 1.6992... Generator Loss: 0.2526
Epoch  1/5 Batch  140/3166... Discriminator Loss: 0.3530... Generator Loss: 5.6309
Epoch  1/5 Batch  150/3166... Discriminator Loss: 0.1150... Generator Loss: 6.5006
Epoch  1/5 Batch  160/3166... Discriminator Loss: 0.2934... Generator Loss: 1.9743
Epoch  1/5 Batch  170/3166... Discriminator Loss: 0.1009... Generator Loss: 4.1225
Epoch  1/5 Batch  180/3166... Discriminator Loss: 2.3695... Generator Loss: 0.1160
Epoch  1/5 Batch  190/3166... Discriminator Loss: 0.1748... Generator Loss: 2.8530
Epoch  1/5 Batch  200/3166... Discriminator Loss: 0.1733... Generator Loss: 2.8576

Epoch  1/5 Batch  210/3166... Discriminator Loss: 0.6192... Generator Loss: 1.0515
Epoch  1/5 Batch  220/3166... Discriminator Loss: 0.0842... Generator Loss: 4.5869
Epoch  1/5 Batch  230/3166... Discriminator Loss: 0.2755... Generator Loss: 2.5110
Epoch  1/5 Batch  240/3166... Discriminator Loss: 0.1926... Generator Loss: 5.3017
Epoch  1/5 Batch  250/3166... Discriminator Loss: 0.2211... Generator Loss: 2.8155
Epoch  1/5 Batch  260/3166... Discriminator Loss: 1.5023... Generator Loss: 0.3546
Epoch  1/5 Batch  270/3166... Discriminator Loss: 0.3131... Generator Loss: 2.1330
Epoch  1/5 Batch  280/3166... Discriminator Loss: 0.3084... Generator Loss: 2.6051
Epoch  1/5 Batch  290/3166... Discriminator Loss: 0.1641... Generator Loss: 4.1145
Epoch  1/5 Batch  300/3166... Discriminator Loss: 1.4274... Generator Loss: 8.6474

Epoch  1/5 Batch  310/3166... Discriminator Loss: 0.1521... Generator Loss: 3.5612
Epoch  1/5 Batch  320/3166... Discriminator Loss: 1.4543... Generator Loss: 0.3955
Epoch  1/5 Batch  330/3166... Discriminator Loss: 0.3572... Generator Loss: 8.8821
Epoch  1/5 Batch  340/3166... Discriminator Loss: 0.2004... Generator Loss: 2.5864
Epoch  1/5 Batch  350/3166... Discriminator Loss: 0.3010... Generator Loss: 2.3100
Epoch  1/5 Batch  360/3166... Discriminator Loss: 0.6695... Generator Loss: 1.1725
Epoch  1/5 Batch  370/3166... Discriminator Loss: 0.1273... Generator Loss: 3.5492
Epoch  1/5 Batch  380/3166... Discriminator Loss: 0.0878... Generator Loss: 6.7740
Epoch  1/5 Batch  390/3166... Discriminator Loss: 0.2903... Generator Loss: 5.2892
Epoch  1/5 Batch  400/3166... Discriminator Loss: 0.8862... Generator Loss: 1.6445

Epoch  1/5 Batch  410/3166... Discriminator Loss: 0.5249... Generator Loss: 6.6362
Epoch  1/5 Batch  420/3166... Discriminator Loss: 1.2290... Generator Loss: 0.5339
Epoch  1/5 Batch  430/3166... Discriminator Loss: 0.2649... Generator Loss: 6.8795
Epoch  1/5 Batch  440/3166... Discriminator Loss: 0.2143... Generator Loss: 6.4961
Epoch  1/5 Batch  450/3166... Discriminator Loss: 0.1740... Generator Loss: 7.2991
Epoch  1/5 Batch  460/3166... Discriminator Loss: 0.2406... Generator Loss: 2.8800
Epoch  1/5 Batch  470/3166... Discriminator Loss: 0.8326... Generator Loss: 0.8172
Epoch  1/5 Batch  480/3166... Discriminator Loss: 4.3370... Generator Loss: 2.4448
Epoch  1/5 Batch  490/3166... Discriminator Loss: 1.1215... Generator Loss: 1.5680
Epoch  1/5 Batch  500/3166... Discriminator Loss: 0.7643... Generator Loss: 1.6494

Epoch  1/5 Batch  510/3166... Discriminator Loss: 1.4988... Generator Loss: 0.7206
Epoch  1/5 Batch  520/3166... Discriminator Loss: 0.6177... Generator Loss: 1.7514
Epoch  1/5 Batch  530/3166... Discriminator Loss: 1.6236... Generator Loss: 2.2372
Epoch  1/5 Batch  540/3166... Discriminator Loss: 0.3944... Generator Loss: 2.5380
Epoch  1/5 Batch  550/3166... Discriminator Loss: 1.0700... Generator Loss: 0.8585
Epoch  1/5 Batch  560/3166... Discriminator Loss: 2.2392... Generator Loss: 0.1897
Epoch  1/5 Batch  570/3166... Discriminator Loss: 0.3488... Generator Loss: 4.2978
Epoch  1/5 Batch  580/3166... Discriminator Loss: 0.2025... Generator Loss: 3.5593
Epoch  1/5 Batch  590/3166... Discriminator Loss: 0.1665... Generator Loss: 4.8634
Epoch  1/5 Batch  600/3166... Discriminator Loss: 0.2524... Generator Loss: 4.5957

Epoch  1/5 Batch  610/3166... Discriminator Loss: 0.4164... Generator Loss: 2.9503
Epoch  1/5 Batch  620/3166... Discriminator Loss: 0.4428... Generator Loss: 1.5454
Epoch  1/5 Batch  630/3166... Discriminator Loss: 0.6595... Generator Loss: 1.5575
Epoch  1/5 Batch  640/3166... Discriminator Loss: 0.3864... Generator Loss: 2.3137
Epoch  1/5 Batch  650/3166... Discriminator Loss: 0.8668... Generator Loss: 5.9272
Epoch  1/5 Batch  660/3166... Discriminator Loss: 1.2259... Generator Loss: 1.2199
Epoch  1/5 Batch  670/3166... Discriminator Loss: 0.8538... Generator Loss: 3.0857
Epoch  1/5 Batch  680/3166... Discriminator Loss: 0.7935... Generator Loss: 0.9516
Epoch  1/5 Batch  690/3166... Discriminator Loss: 0.7630... Generator Loss: 1.2364
Epoch  1/5 Batch  700/3166... Discriminator Loss: 0.7266... Generator Loss: 1.1935

Epoch  1/5 Batch  710/3166... Discriminator Loss: 1.0152... Generator Loss: 0.7378
Epoch  1/5 Batch  720/3166... Discriminator Loss: 2.3418... Generator Loss: 0.9006
Epoch  1/5 Batch  730/3166... Discriminator Loss: 0.3503... Generator Loss: 5.7800
Epoch  1/5 Batch  740/3166... Discriminator Loss: 0.4876... Generator Loss: 1.7141
Epoch  1/5 Batch  750/3166... Discriminator Loss: 0.4926... Generator Loss: 3.2314
Epoch  1/5 Batch  760/3166... Discriminator Loss: 0.4846... Generator Loss: 1.3818
Epoch  1/5 Batch  770/3166... Discriminator Loss: 1.0584... Generator Loss: 0.7560
Epoch  1/5 Batch  780/3166... Discriminator Loss: 0.2283... Generator Loss: 3.3892
Epoch  1/5 Batch  790/3166... Discriminator Loss: 0.1777... Generator Loss: 4.0875
Epoch  1/5 Batch  800/3166... Discriminator Loss: 0.5206... Generator Loss: 3.0098

Epoch  1/5 Batch  810/3166... Discriminator Loss: 0.7702... Generator Loss: 0.9532
Epoch  1/5 Batch  820/3166... Discriminator Loss: 0.5478... Generator Loss: 1.5843
Epoch  1/5 Batch  830/3166... Discriminator Loss: 0.6275... Generator Loss: 5.9460
Epoch  1/5 Batch  840/3166... Discriminator Loss: 1.5219... Generator Loss: 1.8140
Epoch  1/5 Batch  850/3166... Discriminator Loss: 0.4028... Generator Loss: 4.1394
Epoch  1/5 Batch  860/3166... Discriminator Loss: 1.8999... Generator Loss: 0.3523
Epoch  1/5 Batch  870/3166... Discriminator Loss: 0.4338... Generator Loss: 2.0614
Epoch  1/5 Batch  880/3166... Discriminator Loss: 0.2162... Generator Loss: 3.3508
Epoch  1/5 Batch  890/3166... Discriminator Loss: 0.3904... Generator Loss: 2.5255
Epoch  1/5 Batch  900/3166... Discriminator Loss: 0.3937... Generator Loss: 2.5115

Epoch  1/5 Batch  910/3166... Discriminator Loss: 1.0195... Generator Loss: 3.3233
Epoch  1/5 Batch  920/3166... Discriminator Loss: 1.7021... Generator Loss: 2.3684
Epoch  1/5 Batch  930/3166... Discriminator Loss: 0.6126... Generator Loss: 2.3215
Epoch  1/5 Batch  940/3166... Discriminator Loss: 0.6282... Generator Loss: 2.0128
Epoch  1/5 Batch  950/3166... Discriminator Loss: 1.7066... Generator Loss: 0.3935
Epoch  1/5 Batch  960/3166... Discriminator Loss: 0.8219... Generator Loss: 1.5008
Epoch  1/5 Batch  970/3166... Discriminator Loss: 0.7119... Generator Loss: 2.8436
Epoch  1/5 Batch  980/3166... Discriminator Loss: 1.3959... Generator Loss: 0.4595
Epoch  1/5 Batch  990/3166... Discriminator Loss: 0.4515... Generator Loss: 2.0938
Epoch  1/5 Batch 1000/3166... Discriminator Loss: 1.3232... Generator Loss: 3.6344

提交项目¶

提交本项目前，确保运行所有 cells 后保存该文件。

保存该文件为 "dlnd_face_generation.ipynb"，并另存为 HTML 格式 "File" -> "Download as"。提交项目时请附带 "helper.py" 和 "problem_unittests.py" 文件。